Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
by
Staffan Canbäck
March 1997
CONTENTS
1. INTRODUCTION ...................................................................................... 3
2. EXECUTIVE SUMMARY.......................................................................... 4
6. FINDINGS ................................................................................................. 31
8. CONCLUSION ......................................................................................... 36
REFERENCES ..................................................................................................... 37
3
1. INTRODUCTION
State University (Klier 1993). I have chosen to critique this thesis for three
research.
4
2. EXECUTIVE SUMMARY
theory is that there have only been a few attempts at operationalising the
theory’s concepts, and there are even fewer empirical studies. Klier’s
thesis adds to the knowledge in these areas and can usefully serve as a
questionnaire data.
applied. However, the fairly low response rate to the questionnaire and
Klier identifies a crucial problem that has not been researched before, and
earlier research and builds on the traditions of, for example, Monteverde
and Teece (1982), and Masten, Meehan and Snyder (1989) who have
processes.” However, little research has been done on the linkage between
Vertical integration problems are also fertile ground for testing transaction
cost theory. The theory predicts that vertical integration, rather than the
use of market transactions, will be more common when the assets of the
buyer and supplier are linked and interdependent. This so-called asset
specificity can be in the form of site,3 physical asset,4 human asset5 and
buyers and suppliers evolve, being closer to vertical integration than arms-
3 Site specificity arises when successive stages of production need to be located in close proximity
to each other and the set-up or relocation costs are great.
4 Physical asset specificity is high if assets are dedicated to a specific purpose and can not easily be
used in other operations.
5 Human asset specificity exists if human resources are trained specifically for a stage in the
production process and the skill cannot be used elsewhere.
6 Dedicated asset specificity exists when a supplier expands capacity based on demand from a
particular buyer. Klier does not refer to dedicated asset specificity.
8
twenty years.
Klier thus addresses an important issue and his research passes any test of
Both the problem and the hypothesis are delimited and narrow since they
problems or hypotheses that should have been tested at the same time.
The problem and hypothesis are well linked to the theoretical foundation.
7 Klier's hypothesis can be misunderstood since a decrease in vertical integration might lead the
reader to believe that there will be a commensurate increase in market-based relationships (no
integration). However, the proportion of such relationships will also decline according to Klier's
hypothesis. The difference is made up of quasi-integration.
10
without testing its validity, or even recognising that the validity can be
market over the last 20 years, while vertical integration has decreased.
vertical integration. Walker and Weber (1984) made the same observation
3.4 RESEARCHABILITY
The problem is researchable since Klier breaks down his hypothesis into
8 A useful distinction may be to divide uncertainty into at least two components: volatility
uncertainty and variety uncertainty. Volatility refers to temporal fluctuations, variety to degrees
of freedom in technological and market choices.
11
are aligned with earlier research so that older data can be used to support
Klier also recognises the limitations of his research. For example, he states
4. LITERATURE REVIEW
The literature survey is thorough and to the point. It covers the relevant
Klier chooses to divide the literature survey in two parts. First he discusses
and how the paradigm shift should affect vertical integration if transaction
focuses on the key aspects of relevance to the research and leaves out
4.1.1 Fordism
Klier summarises the key aspects of the Fordist system on pages 13–17. He
suppliers.
Cohen and Zysman (1987) are referenced on the buffer tendency, Scherrer
(1991) is used to prove the vertical integration point, and Womack, Daniel
and Roos (1990) and the MIT Commission (1989) are used as evidence for
Klier’s view on how the Fordist system works and this section of the
literature review is appropriate in length and scope. The only issue may be
contrast it with the Fordist system. His conclusion is that the new system
Klier. From this stems a number of practices that differ sharply from the
Fordist system. Teamwork is at the core of the system and the separation
used to show that the value added sales ratio in Japan (just-in-time) is 15–
20 per cent, while the North American (Fordist) ratio was 36–48 per cent
explain why this paradigm is winning? Womack, Daniel and Roos (1990)
Finally, Klier discusses the implications of the paradigm shift on pages 25–
increase when just-in-time systems are implemented. The reason for this is
that the asset specificity (see Section 4.2, below, on transaction costs)
make components and sub-assemblies in-house. Klier then argues that this
is not happening because OEMs and their suppliers are entering into new
increased from 1.2 years in 1984 to 2.3 years in 1989. Furthermore, the
most important purchasing criteria changed from price to quality over the
same time period. Womack, Daniel and Roos (1990) demonstrate how the
studied the locational patterns in the United States and found that
This overview of the changes is compelling, but Klier does not explain
why the changes are happening, only that they are happening. In fact,
would have made a better argument if he had referenced this work as well
and the first part of Chapter 3, in total 27 pages covering the underlying
4.2.1 Theory
This could have been a lengthy discussion because much has been written
on the theory over the last 20 years, and vertical integration problems have
been at the core of this effort. Fortunately, Klier summarises the key
and other definitions exist. Klier should have referred to this and then
(1984) makes a similar point within the transaction cost framework. In the
end, Klier would probably have come to the same choice he did since most
empirical work use the definition above, but his choice would have
studies that address the issue of vertical integration within contexts similar
potentially could have been relevant to his research and explains why this
Pages 36–54 review the ten studies. Among these ten studies are the
widely quoted ones by Monteverde and Teece (1982), Masten (1984) and
19
Masten, Meehan and Snyder (1989, 1991). Each of these operationalises the
and tests their explanatory power in a static setting. Two of the studies are
from the automotive industry. Thus, they are highly appropriate to Klier
and the only difference to his research is that he is testing for dynamic
effects.
and notes that while their methodology was sound their sample was
small. Thus, the significance of their test was not strong. Unfortunately,
this comment is not carried forward into the statistical part of the thesis
Klier also reviews Masten, Meehan and Snyder (1991) extensively and
notes that their study not only tested under what conditions a firm
impact of these choices. It could have been worthwhile to discuss why this
approach was not appropriate in Klier’s study, especially since this study
has been cited as the most interesting empirical study of transaction costs
(Joskow 1991).
20
Finally, a notable omission from the review are the two studies in the
automotive industry by Walker and Weber (1984, 1987) who focus on the
Monteverde and Teece in their first study and make the important
They find that high demand uncertainty leads to vertical integration while
approach.
not material.
21
5. ANALYTICAL APPROACH
First, he needs to show that the preferred governance model has changed
validation of the quality of the data nor a discussion of what would have
Klier chose to test his hypothesis by collecting data from the three
of engineering effort”.
The definitions are thus fairly loose and open to a wide range of
was the only way to get information from the car manufacturers, but the
Klier also had to draw on data from earlier research to complement his
current industry survey. However, earlier studies have not tested Klier’s
find ways to extract information that helped him test his hypothesis. Both
Monteverde and Teece (1982) and Masten, Meehan and Snyder (1991)
23
provided such data. Unfortunately, data from the most closely related
study in the automotive industry (Masten, Meehan and Snyder 1989) was
Despite the fact that this earlier data was quite different from what he
comparisons difficult.
While there are limitations in the choice of research methodology, and the
Klier clearly explains why he chose the methods used, discusses the merits
and problems of each choice made, and applies the statistical methods
rigorously.
rather than linear regression models which are intended for continuous
Teece 1982; Masten, Meehan and Snyder 1989) since he does not have
independent variables in his own study with the previous studies. Klier
There are a few problems with this approach. First, Klier has to collapse
“quasi-integration” and “no integration” into one state since the earlier
studies did not make the distinction between these two states. This
Second, even though he states that the two earlier studies used a binomial
probit method, only Monteverde and Teece did. Masten, Meehan and
Snyder use OLS, two-limit tobit, and logit methods. It is unclear from the
Third, there is confusion about the data source for marginal effects for
Masten, Meehan and Snyder. Klier is using their 1991 study of a naval
Monteverde and Teece data and to expand the data set used when
comparing with Klier. The original article uses the independent variable
and Teece’s data can be appended to fit Klier’s more rigorous definition.
The second part of Klier’s statistical analysis aims at showing whether the
between vertical parties. The choice of models for this part of the research
is not controversial. The three states are clearly ordered and the use of an
uses two multinomial choice models: ordered probit and sequential probit.
These models allow Klier to use three states in the dependent variable:
that he can thus estimate to what extent just-in-time techniques, with its
predicted in his hypothesis. The ordered probit model is used under the
assumption that the buyer and supplier decide at one point in time which
governance model to use. The sequential probit model assumes that a first
choice is made between internal and market transaction. Only later, in the
and no integration.
27
and 2 for vertical integration, b' is the coefficient vector and x is the
allowed them to incorporate cost data for those parts that were made
collect similar data for external parts. The approach makes it possible to
Clearly, Klier does not have access to such data from the OEMs he
surveyed, and it is unlikely that he can get it. This approach suggests a
forward. Perhaps Klier should also have made reference to this approach
in his thesis.
specificity as either due to the car’s style or due to the technical design.
and the plant of the auto assembler, in line with Spiller (1985).
identifies four other proxies for commitment: the average number of firms
The statistical methods are applied rigorously and all key results are
nothing to ask for. Klier also tests the validity of the results and
The results are also compared with Monteverde and Teece (1982) and
Masten, Meehan and Snyder (1991) and are found to correlate fairly
6. FINDINGS
line with his hypothesis. Thus, he argues that the strong anecdotal
evidence built over the previous decade has been validated by his
statistical significance of the findings, and the quality of the data, to create
a complete picture.
alternatives, it is still likely that Klier would have picked the same
and one can argue that frequency of delivery is nothing else than
frequency of delivery. The other proxies are more solid, but also less
important.
(p. 99). However, this analysis is not optimal since it does not explicitly
time. The ordered probit and sequential probit analyses are less significant
compare to earlier studies, and the sample size is not wholly adequate.
On balance, Klier’s findings are not robust but they do add insights on the
margin. The results are in line with earlier research and contribute to the
33
One avenue for this is to replicate Klier’s study but with a different set of
the economic impact, and the granularity of the analysis can be improved.
Perhaps this particular aspect of the vertical integration problem has been
exhausted, though.
34
7. THESIS LAYOUT
The thesis covers 119 pages (excluding two short appendices) and roughly
The mechanics applied are excellent. Klier uses good academic grammar,
there are few—if any—spelling errors, and quotes and footnotes are
correct and relevant. The reference list is mostly accurate, although the
Econometrica), and both Eckard (1984) and Masten, Meehan and Snyder
chapters, headings, etc., since one typeface and font size is uniformly
applied.
overly detailed (if the aim is to give a summary description) or too short (if
On balance, though, the power and clarity of the language makes the
8. CONCLUSION
hypothesis. The whole thesis is targeted at this task and there is little
excess discussion. The findings are in line with earlier findings although
implications.
operational.
37
REFERENCES
Coase, R. H. 1937. The Nature of the Firm. Economica n.s., 4 (16): 386–405.
———. 1985. The Economic Institutions of Capitalism. New York: Free Press.
Womack, J. P., D. T. Jones and D. Roos. 1990. The Machine That Changed the
World. New York: Rawson.